Re: undo make-bcache (was: Re: Can't mount an encrypted backing device)

[Coly Li <colyli@suse.de>]

On 2020/1/18 9:43 下午, Clodoaldo Neto wrote:
> On Sat, Jan 18, 2020 at 9:35 AM Coly Li <colyli@suse.de> wrote:
>>
>> On 2020/1/18 8:22 下午, Clodoaldo Neto wrote:
>>> On Sat, Jan 18, 2020 at 7:54 AM Clodoaldo Neto
>>> <clodoaldo.pinto.neto@gmail.com> wrote:
>>>>
>>>> On Thu, Jan 16, 2020 at 9:59 PM Coly Li <colyli@suse.de> wrote:
>>>>>
>>>>> On 2020/1/17 5:52 上午, Clodoaldo Neto wrote:
>>>>>>
>>>>>> Em seg, 13 de jan de 2020 11:19, Coly Li <colyli@suse.de
>>>>>> <mailto:colyli@suse.de>> escreveu:
>>>>>>>
>>>>>>> On 2020/1/13 8:44 下午, Jens-U. Mozdzen wrote:
>>>>>>>> Hi Coly,
>>>>>>>>
>>>>>>>> jumping in here, because I was looking for a way to revert from bcache
>>>>>>>> to plain device:
>>>>>>>>
>>>>>>>> Zitat von Coly Li <colyli@suse.de <mailto:colyli@suse.de>>:
>>>>>>>>> The super block location of the backing disk is occupied by bcache. You
>>>>>>>>> cannot mount the file system directly from the backing disk which is
>>>>>>>>> formated as bcache backing device [...] (bcache offset all I/Os on
>>>>>>>>> bcache device 4KB behind the requesting
>>>>>>>>> LBA on backing disk).
>>>>>>>>
>>>>>>>> Assuming that no caching device is associated with a backing device (so
>>>>>>>> the backing device is "clean" as in "containing all data blocks with the
>>>>>>>> current content"), could one convert the content of a backing device to
>>>>>>>> a "non-bcached device" by removing the first 4096 octets of the backing
>>>>>>>> device content?
>>>>>>>>
>>>>>>>> Something like "dd if=backingdev of=newdev skip_bytes=4096 ..."?
>>>>>>>
>>>>>>> Hi Jens-U,
>>>>>>>
>>>>>>> you may try dmsetup to setup a linear device mapper target, and the map
>>>>>>> table just skipping the first 4KB (bcache superblock area). If you are
>>>>>>> lucky, I mean the real file system is not corrupted, the created device
>>>>>>> mapper target can be mounted directly.
>>>>>>
>>>>>>
>>>>>> I'm trying dmsetup but it does not accept anything other than 0 and 0
>>>>>> at the beginning and end of the table:
>>>>>>
>>>>>> # echo '0 3774578672 linear /dev/mapper/backing-device 8' | dmsetup
>>>>>> create dmb
>>>>>> device-mapper: reload ioctl on dmb  failed: Invalid argument
>>>>>> Command failed.
>>>>>
>>>>> The above line should work, if 3774578672 is a correct size number in
>>>>> sectors.
>>>>
>>>> I took it from the original map:
>>>>
>>>> # dmsetup table /dev/mapper/backing-device
>>>> 0 3774578672 crypt aes-xts-plain64
>>>> :64:logon:cryptsetup:7e2c0b40-8dec-4b13-8d00-b53b55160775-d0 0 251:0
>>>> 32768
>>>
>>> It works like this:
>>>
>>> # echo '0 3774578664 linear /dev/mapper/backing-device 8' | dmsetup create dmb
>>>
>>> But then I can't mount it:
>>>
>>> # mount /dev/mapper/dmb /r
>>> mount: /r: wrong fs type, bad option, bad superblock on
>>> /dev/mapper/dmb, missing codepage or helper program, or other error.
>>
>> It might be my fault, from bcache-tools, it seems the offset is
>> BDEV_DATA_START_DEFAULT (16 sectors). How about:
>> # echo '0 3774578656 linear /dev/mapper/backing-device 16' | dmsetup
>> create dmb
> 
> Still no luck
> 
> # echo '0 3774578656 linear /dev/mapper/backing-device 16' | dmsetup create dmb
> # mount /dev/mapper/dmb /r
> mount: /r: wrong fs type, bad option, bad superblock on
> /dev/mapper/dmb, missing codepage or helper program, or other error.

The tricky part is to calculate the correct linear mapping size.

For example, I have a 500G hard drive, from fdisk I see its size is
1048576000 sectors. I use it as a bcache backing device and format an
Ext4 file system on it. Then I use the following mistaken command line
to setup a linear target to skip the first 8 sectors,
 # echo '0 1048575000  linear /dev/sdc 16' | dmsetup create dmb
Then mount /dev/mapper/dmb to /mnt fails, the kmesg hints me,
 [1884572.477316] EXT4-fs (dm-0): bad geometry: block count 131071998
exceeds size of device (131071875 blocks)

So I realize Ext4 file system will check the block device size, so I
need to provide an exact accurate length number for the linear target.
From the kmesg I see the correct sector size should be
(131071998*8=)1048575984, then I re-create the linear target by,
 # echo '0 1048575984  linear /dev/sdc 16' | dmsetup create dmb
Then I mount /dev/mapper/dmb to /mnt, it works and I have the following
line from command 'mount',
 /dev/mapper/dmb on /mnt type ext4 (rw,relatime)

I don't use the encrpyt file system and I don't know the excact detail
about it. But I guess, it might because the incorrect length number for
your dm linear target as I show by the above example.

-- 

Coly Li